Rapid fingerprint identification of cannabis-based drugs via terahertz time-domain spectroscopy and machine learning

We present a non-destructive, non-contact method for rapid identification of cannabis-derived substances using terahertz time-domain spectroscopy combined with machine learning. Spectral data were acquired from seven cannabis sample types across 0.1-2.0 THz. Principal component analysis coupled with a linear discriminant analysis classifier was applied to a fused feature vector of absorption coefficient, refractive index, and phase-difference spectra, yielding a mean cross-validated accuracy of 98.3% ± 1.1% and a hold-out test accuracy of 90.5%. Characteristic absorption peaks were observed at 0.62 THz (THC) and 0.58 THz (CBD). The complete identification process takes less than five minutes, approximately eight times faster than GC-MS, and penetrates non-metallic packaging without contact. Five non-cannabis substances were clearly separated in the PC1-PC2 score space, confirming specificity. This approach provides an efficient, safe solution for on-site forensic screening of cannabis products.